 |
Paper accepted at 2019 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) |
|---|
 |
 |
 |
 |
 |
 |
 |
|---|---|---|---|---|---|---|
| Junting Pan | Chengyu Wang | Xu Jia | Jing Shao | Lu Sheng | Junjie Yan | Xiaogang Wang |
This paper proposes the novel task of video generation conditioned on a SINGLE semantic label map, which provides a good balance between flexibility and quality in the generation process. Different from typical end-to-end approaches, which model both scene content and dynamics in a single step, we propose to decompose this difficult task into two sub-problems. As current image generation methods do better than video generation in terms of detail, we synthesize high quality content by only generating the first frame. Then we animate the scene based on its semantic meaning to obtain the temporally coherent video, giving us excellent results overall. We employ a cVAE for predicting optical flow as a beneficial intermediate step to generate a video sequence conditioned on the initial single frame. A semantic label map is integrated into the flow prediction module to achieve major improvements in the image-to-video generation process. Extensive experiments on the Cityscapes dataset show that our method outperforms all competing methods.
Find our work on arXiv.
Please cite with the following Bibtex code:
@article{pan2019video,
title={Video Generation from Single Semantic Label Map},
author={Pan, Junting and Wang, Chengyu and Jia, Xu and Shao, Jing and Sheng, Lu and Yan, Junjie and Wang, Xiaogang},
journal={arXiv preprint arXiv:1903.04480},
year={2019}
}
You may also want to refer to our publication with the more human-friendly Chicago style:
Junting Pan, Chengyu Wang, Xu Jia, Jing Shao, Lu Sheng, Junjie Yan and Xiaogang Wang. "Video Generation from Single Semantic Label Map." CVPR 2019.
The Seg2Vid presented in our work can be downloaded from the links provided below the figure:
Seg2Vid Architecture
Img2Vid Architecture
 |
 |
|---|
 |
 |
|---|---|
 |
 |
 |
 |
 |
|---|
 |
 |
 |
 |
 |
 |
|---|
 |
 |
 |
 |
 |
|---|
-
Cityscapes
- Cityscapes dataset can be downloaded from the official website (registration required).
- We apply Deeplab-V3 github-repo to get the corresponding semantic maps.
- We organize the dataset following as below:
seg2vid ├── authors ├── figs ├── gifs ├── logos ├── pretrained_models ├── src ├── data │ ├── cityscapes │ │ ├── leftImg8bit_sequence │ │ │ ├── train_256x128 │ │ │ │ ├── aachen │ │ │ │ │ ├── aachen_000003_000019_leftImg8bit.png │ │ │ ├── val_256x2128 │ │ │ ├── val_pix2pixHD │ │ │ │ ├── frankfurt │ │ │ │ │ ├── frankfurt_000000_000294_pix2pixHD.png │ │ │ ├── train_semantic_segmask │ │ │ ├── val_semantic_segmask │ │ │ │ ├── frankfurt │ │ │ │ │ ├── frankfurt_000000_000275_ssmask.png │ │ ├── gtFine │ │ │ ├── train │ │ │ ├── val │ │ │ │ ├── frankfurt │ │ │ │ │ ├── frankfurt_000000_000294_gtFine_instanceIds.png -
KTH
- We use the KTH human action dataset dataset, and we follow the data processing in svg.
-
UCF-101
- UCF-101 dataset can be downloader from the official website
python -u test_refine_w_mask_two_path.py --suffix refine_w_mask_two_path --dataset cityscapes_two_path
python -u train_refine_multigpu_w_mask_two_path.py --batch_size 8 --dataset cityscapes_two_path
Seg2Vid is implemented in Pytorch.
If you have any general doubt about our work or code which may be of interest for other researchers, please use the public issues section on this github repo. Alternatively, drop us an e-mail at mailto:junting.pa@gmail.com.